[jira] [Commented] (SPARK-14141) Let user specify datatypes of pandas dataframe in toPandas()

["Luke Miner (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-14141?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15830750#comment-15830750 ] 

Luke Miner commented on SPARK-14141:
------------------------------------

One option is to convert all the categorical variables into integers in spark and then once we have the pandas dataframe, re-encode them as categoricals. Here's a dummy example.

{code}
def to_pandas_categorical(df, col_names):
    def encode_string_with_int(df, col_names):
        column_codes = {}
        for col_name in col_names:
            # get encodings for later use by pandas
            unique_values = (df
                             .select(col_name)
                             .distinct()
                             .rdd
                             .map(lambda r: r[0])
                             .collect())
            encodings = map(str, range(len(unique_values)))
            column_codes[col_name] = unique_values

            # now encode df column
            df = df.replace(unique_values, encodings, col_name)
            df = df.withColumn(col_name, F.col(col_name).cast('integer'))
        return df, column_codes

    df_encoded, column_codes = encode_string_with_int(df, col_names)
    df_encoded_pdf = df_encoded.toPandas()
    for col_name in col_names:
        df_encoded_pdf[col_name] = pd.Categorical.from_codes(df_encoded_pdf[col_name],
                                                             column_codes[col_name])
{code}

> Let user specify datatypes of pandas dataframe in toPandas()
> ------------------------------------------------------------
>
>                 Key: SPARK-14141
>                 URL: https://issues.apache.org/jira/browse/SPARK-14141
>             Project: Spark
>          Issue Type: New Feature
>          Components: Input/Output, PySpark, SQL
>            Reporter: Luke Miner
>            Priority: Minor
>
> Would be nice to specify the dtypes of the pandas dataframe during the toPandas() call. Something like:
> bq. pdf = df.toPandas(dtypes={'a': 'float64', 'b': 'datetime64', 'c': 'bool', 'd': 'category'})
> Since dtypes like `category` are more memory efficient, you could potentially load many more rows into a pandas dataframe with this option without running out of memory.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org